Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Last revisionBoth sides next revision
doku:mpi [2017/03/23 14:44] – [VSC-2] irdoku:mpi [2021/09/14 13:27] – [Process Pinning with MPI] goldenberg
Line 12: Line 12:
  
  
-==== Process Pinning with MPI ==== 
-Most MPI implementations allow a pinning of the MPI processes to specific CPU cores if needed. 
- 
- 
-Numbering of the cores on a VSC-1 node with 8 cores: 
-<code> 
-CPU0 CORE0 
-CPU0 CORE1 
-CPU0 CORE2 
-CPU0 CORE3 
- 
-CPU1 CORE4 
-CPU1 CORE5 
-CPU1 CORE6 
-CPU1 CORE7 
-</code> 
- 
-Numbering of the cores on a VSC-2 node with 16 cores: 
-<code> 
-CPU0 CORE0 
-CPU0 CORE1 
-CPU0 CORE2 
-CPU0 CORE3 
-CPU0 CORE4 
-CPU0 CORE5 
-CPU0 CORE6 
-CPU0 CORE7 
- 
-CPU1 CORE8 
-CPU1 CORE9 
-CPU1 CORE10 
-CPU1 CORE11 
-CPU1 CORE12 
-CPU1 CORE13 
-CPU1 CORE14 
-CPU1 CORE15 
- 
-</code> 
- 
- 
-=== Pinning to cores 0,1,2,4,6 === 
- 
-Mvapich: 
-<code> 
-mpirun  -machinefile $TMPDIR/machines -np 1 -env MV2_CPU_MAPPING 0-2,4,6 ./a.out  
-</code> 
- 
- 
-Intel MPI: 
-<code> 
-export I_MPI_PIN_PROCESSOR_LIST=0-2,4,6 
-mpirun -machinefile $TMPDIR/machines -np 1 ./a.out 
-</code>